Accurately Identifying Cause and Effect in PPC Marketing.


What is Cause and Effect?

You have undoubtedly heard of cause and effect. Cause and effect (or causality) is identifying the relationship between two events, specifically in how the second event is related to the first. Did the first event cause the second, or is there another unknown event involved?

Wikipedia definition of causality

Freakonomics: The Cause & Effect of Ice Cream

One of my favorite books is Freakonomics. I’m not sure if all marketers like it as much as I do, but it was revolutionary for me in two key ways. First, it helped me learn to think outside of the box with the data I look at day after day. It caused me to stop accepting the easy answer, and to start digging for the right answer… even if sometimes those two were the same thing! The second thing the book did for me, was to challenge my amateur understanding of the nature of cause and effect. The following video is a brief excerpt from the Freakonomics movie (currently available on Netflix streaming) on their “Cause and Effect” section.

How does Cause and Effect work into PPC Marketing?

In many ways, accurately identifying the relationship of cause and effect in PPC centers around analyzing optimization changes.

Simplistic Scenario:

You are bidding on the keyword [subaru for sale]. You are only getting 100 impressions and 1 click per day on your ads and your avg ad position is 8.5 so you decide to raise your keyword bid from $2.50 to $3.50. You check this keyword the next day and see that your impressions have gone up to 200 and your clicks have gone up to 5. You, therefore, (consciously or not) make the assumption that the cause of the effect (more impressions and clicks) was this: you raised your KW bid. Therefore, like every good marketer, you run off and raise all your KW bids so you can increase your traffic (kidding, don’t do that!).

Here’s the problem, even in this simplistic scenario it can be difficult to truly identify the actual cause and effect relationship. Unless you have an accurate handle on the various factors that can affect your client and your account, you could spend all your time chasing a false “cause” for the effect that you identified.


Realistic Scenario:

Here’s a bit of a more realistic (and complex) scenario that a PPC account manager will run into. You are doing a monthly report for a client with unlimited budget and notice that traffic, cost, revenue, and conversion rates have all risen since the previous month, but thankfully so has ROI! You are sitting well and send a glowing report to the client who thanks you for your hard work. You immediately do a happy dance and then as a good marketer refuse to stop there, but instead dig in to identify the reason for this increase. You start digging through Change History in an attempt to identify which of your optimizations resulted in such a fantastic increase in happiness.

Unfortunately, the more you dig, the more confused you get. You think it could have something to do with some new ad A/B tests you ran, but then you see that BOTH the B ads and the A ads demonstrated a higher quality of traffic and sales. You dig around some more wondering if various keyword exclusions and bid adjustments led to the rise. It seems that basic optimizations had some level of role in the better numbers, but they don’t provide a clear explanation for the changes. Honestly, the more you dig the more stumped you are so you decide to give the client a call to see if you’re missing something. You talk for awhile, and at the end of the call she happens to mention something, “Oh, I’m not sure if this is helpful but we actually lowered our prices by $.50 at the beginning of the month to be more competitive with some new players in the industry.” All of a sudden, it begins to fall in place. You had been focused entirely on the “causal” optimizations you made in the account, and completely left out the more significant “cause” of a lower price in SERPs. Since you spend a lot on Google Shopping, you realize that this lower price has jumped out even more to potential customers in your highly competitive market, thus more traffic and more sales.

Possible Causation Factors

It is essential in PPC, to not merely be satisfied with the initial “cause” but to be willing to do the work to dig more when needed. Obviously, there are scenarios more like the Simplistic Scenario above where the factors involved point more readily and accurately to the most obvious cause. However, in PPC, it is essential that we always keep in mind the various factors that can be causes (even if only minor) of the effect we are observing. Here are several factors I could think of that should be considered when analyzing PPC data. Did I miss any? Leave your thoughts in the comments below.

  • Seasonality (Do past years demonstrate a similar change in behavior?)
  • Product Pricing (For e-com sites, price changes can significantly impact behavior.)
  • Product Sales (Is the client running a discount or deal?)
  • Competitor Pricing Changes (For e-com sites, has your competition recently changed their prices that would help explain a change in behavior?)
  • sale in ppc products

  • Competition (Is there new competition in your space raising your bids? Did competition drop out or hit a budget cap to help explain an inexplicable bump in behavior?)
  • Bidding (Have you made bidding changes around the time of your observed effect that would explain the different behavior?)
  • Quality Score Changes (Since QS affects your cost, have there been any recent changes in QS?)
  • Ad Creative Changes (Are you running tests on ads that would help explain a change in user behavior?)
  • Website Changes (Have there been any changes to the client website that would explain the change in user behavior? If it is an e-com client, make sure to double-check the checkout process!!!!)
  • Landing Page Changes (Have there been any changed in the LP CTA, LP design, or other factors that could have affected user behavior?)
  • …and so many more…

How to Accurately Identify Causation in PPC

So after being overwhelmed by the sheer impossibility of 100% accurately identifying the exact cause of every effect, how can we actually optimize with the information we see? As you probably know, it’s pretty unlikely that any effect has one solo cause… which merely complicates everything! I’m curious to know other strategies on this, but I think this is where simplification, repeatability, and common sense are essential.

  1. Simplification – If you are going to have any prayer of identifying the cause (or most significant cause) narrowing down the dataset to the most relevant data for the decision is essential. Use date ranges to compare when the main changes in campaigns/ad groups/keywords began/end. Use filters in your Adwords/Analytics to get rid of the data that has not really changed so you can more easily analyze the data that has changed. There are lots of other ways to better analyze the data, so do what you can to narrow your dataset down to make sure you are actually looking at the data from which you should make your causal decision.
  2. Repeatability – This is pretty much Science 101, but if an observable action is reproducible then there is a good indication that it was the right action. This takes time and unfortunately, the influence of time itself in PPC changes even the most similar action so pure repeatability is not entirely possible. However, we can still make accurate assumptions based upon the most significant data and then reproduce that. In PPC, this is also proven by a continuation of the change. If there is a single bump or jump that levels back down, it’s a good indication this is probably more of an anomaly. If the change causes a steady increase/decrease in the desired behavior, then it’s a good indication that the cause was legitimate.
  3. Common Sense – At the end of the day, the PPC manager (like all analysts) has to make an informed decision based upon an assumption. This is, of course, why it is absolutely essential to hire well, whether agency or in-house. If an account manager knows how to find the data, but doesn’t know what to do with it than that’s not very helpful either :/.

Hopefully this has been an article that has enticed you to think more about the relationship of cause and effect in PPC Marketing. What about you? What are common mistakes you see PPC managers making, or what are some helpful tips you have for us? Just tweet them to me at @PPCKirk.

And enjoy this cartoon.
PPC Marketing correlation and causation